Skip to content

v15.15.0

Latest
Latest

Choose a tag to compare

View all tags
@sickn33 sickn33 released this 18 Aug 07:31
v15.15.0
746ad69
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

[15.15.0] - 2026-08-18 - "Evidence, Durable Context, and UI Contracts"

Added six focused skills for deep reading, durable agent context, measurable
UI work, evaluation reporting, and cross-platform contract audits, while
hardening exact-release installation and contributor-credit workflows. The
published catalog contains 2,019 skills.

This release helps Claude Code, Cursor, Codex CLI, Gemini CLI, Antigravity, and
related AI coding assistants keep claims attached to evidence, preserve verified
project knowledge across sessions, evaluate agent outcomes without denominator
drift, and carry product contracts consistently from storage to clients.

Start here:

Added

  • Added anti-ui-slop for grounding web and iOS work in
    real interface references, explicit product contracts, required interaction
    states, and a finish gate that rejects generic coding-agent defaults
    (#1166).
  • Added liuguang-banlan-ui for two parameterized
    iridescent UI modes with OKLCH authoring, deterministic WebGL and CSS fallback,
    reduced-motion handling, screenshot QA, and measurable color reports
    (#1154).
  • Added using-lwc for durable, source-grounded agent memory
    across Wiki document and CodeGraph planes, with explicit scope, installation,
    initialization, and write-consent boundaries
    (#1152).
  • Added dsh-deepread for evidence-first reading of
    articles, books, PDFs, web pages, and document sets through claim ledgers,
    argument analysis, knowledge maps, and Feynman checks
    (#1174).
  • Added agent-evaluation-reporting for
    keeping autonomous, assisted, failed, timed-out, and invalid evaluation
    outcomes distinct, with explicit denominators, uncertainty, and readiness
    gates (#1177).
  • Added
    cross-platform-contract-propagation-audit
    for read-only tracing of fields, enums, flags, and API contracts across
    storage, services, clients, analytics, rollout controls, and tests
    (#1178).

Changed

  • Updated the exact-release installer to accept npm 12's valid single-element
    JSON-array response for npm view --json, while rejecting empty or ambiguous
    multi-item results and preserving fail-closed gitHead verification
    (#1167).
  • Hardened zero-diff contributor-credit intake so an authoritative empty raw Git
    diff can proceed without weakening the rejection of missing or non-empty
    evidence
    (#1171,
    #1172,
    #1173).
  • Regenerated the canonical catalog, offline AAS Core data, web assets,
    marketplaces, editorial bundles, compatibility reports, and Codex/Claude
    plugin distributions for 2,019 skills.

Security

  • Preserved exact npm package-to-Git identity checks across npm 12 output-shape
    changes; missing, empty, changed, or ambiguous registry identity remains a
    hard installer failure.
  • Repaired the protected maintainer intake for liuguang-banlan-ui and
    using-lwc with explicit provenance, risk labels, consent gates, pinned LWC
    bootstrap identities, reduced-motion behavior, renderer fallback, and
    regression coverage
    (#1169).
  • Kept optional UIZZE preview inspection behind minimized-payload review and
    explicit network approval, and kept LWC installation, global initialization,
    and durable writes behind current user authorization.

Who should care

  • Researchers, students, and analysts who need long-form reading outputs whose
    claims remain linked to inspectable source evidence.
  • Coding teams that need verified project decisions and code relationships to
    survive across agent sessions without silently widening workspace authority.
  • UI teams that want either a strict anti-generic finish gate or a measurable,
    accessible iridescent rendering workbench.
  • Agent-evaluation owners and cross-platform product teams that need honest
    denominators, explicit readiness gates, and evidence for every contract edge.
  • Maintainers and npm 12 users who depend on exact-release identity and
    contributor-credit workflows that fail closed on ambiguous evidence.

Validation

  • Passed repository validation, reference validation, documentation-security
    checks, warning-budget enforcement, the complete 111-group repository test
    suite, plugin-compatibility and bundle checks, web-app install/build/prerender,
    and the npm package dry run on the protected release base.
  • Reviewed all six added skills and their bundled files for semantics, safety,
    provenance, declared risk, limitations, consent boundaries, and relevant
    regression coverage.

Limitations

  • dsh-deepread requires readable source material and cannot recover or invent
    content that the host agent could not retrieve.
  • using-lwc documents LWC workflows but does not bundle an initialized Wiki,
    trusted project scope, or permission to install software or write memory.
  • The UI skills do not guarantee visual quality automatically: external preview
    use needs approval, screenshot claims need actual image inspection, and WebGL
    environments still require the documented fallback and accessibility checks.
  • Evaluation reports and contract audits expose evidence gaps; they do not make
    missing data comparable or implement the repairs they identify.

Credits

Assets 2
Loading